Performance of a quantum annealer on range-limited constraint satisfaction problems

TL;DR

This study evaluates a quantum annealer's performance on specially designed constraint satisfaction problems with limited coupling strength, revealing a potential quantum speedup and highlighting the impact of analog control errors.

Contribution

It introduces a new class of constraint satisfaction instances with limited coupling range and demonstrates a scaling advantage of the quantum annealer over classical algorithms.

Findings

Quantum annealer shows scaling advantage on new problem instances.

Performance is affected by analog control errors.

Potential for quantum speedup as hardware improves.

Abstract

The performance of a D-Wave Vesuvius quantum annealer was recently compared to a suite of classical algorithms on a class of constraint satisfaction instances based on frustrated loops. However, the construction of these instances leads the maximum coupling strength to increase with problem size. As a result, larger instances are subject to amplified analog control error, and are effectively annealed at higher temperatures in both hardware and software. We generate similar constraint satisfaction instances with limited range of coupling strength and perform a similar comparison to classical algorithms. On these instances the D-Wave Vesuvius processor, run with a fixed 20$\mu$s anneal time, shows a scaling advantage over the software solvers for the hardest regime studied. This scaling advantage opens the possibility of quantum speedup on these problems. Our results support the hypothesis that performance of D-Wave Vesuvius processors is heavily influenced by analog control error, which can be reduced and mitigated as the technology matures.